ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.2 by root, Sat May 10 23:33:40 2008 UTC vs.
Revision 1.71 by root, Fri Jun 10 12:35:18 2011 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
45
46#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE
48#endif
2#include "xthread.h" 49#include "xthread.h"
3 50
4#include <errno.h> 51#include <errno.h>
5
6#include "EXTERN.h"
7#include "perl.h"
8#include "XSUB.h"
9
10#include <stddef.h> 52#include <stddef.h>
11#include <stdlib.h> 53#include <stdlib.h>
54#include <string.h>
12#include <errno.h> 55#include <errno.h>
13#include <sys/types.h> 56#include <sys/types.h>
14#include <sys/stat.h> 57#include <sys/stat.h>
58#include <sys/statvfs.h>
15#include <limits.h> 59#include <limits.h>
16#include <fcntl.h> 60#include <fcntl.h>
17#include <sched.h> 61#include <assert.h>
62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
18 68
19#ifndef EIO_FINISH 69#ifndef EIO_FINISH
20# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 70# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
21#endif 71#endif
22 72
… …
29#endif 79#endif
30 80
31#ifdef _WIN32 81#ifdef _WIN32
32 82
33 /*doh*/ 83 /*doh*/
34
35#else 84#else
36 85
37# include "config.h"
38# include <sys/time.h> 86# include <sys/time.h>
39# include <sys/select.h> 87# include <sys/select.h>
40# include <unistd.h> 88# include <unistd.h>
41# include <utime.h> 89# include <utime.h>
42# include <signal.h> 90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif
105
106#ifdef _D_EXACT_NAMLEN
107# undef D_NAMLEN
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif
43 114
44# ifndef EIO_STRUCT_DIRENT 115# ifndef EIO_STRUCT_DIRENT
45# define EIO_STRUCT_DIRENT struct dirent 116# define EIO_STRUCT_DIRENT struct dirent
46# endif 117# endif
47 118
48#endif 119#endif
49 120
50# ifndef EIO_STRUCT_STAT
51# define EIO_STRUCT_STAT struct stat
52# endif
53
54#if HAVE_SENDFILE 121#if HAVE_SENDFILE
55# if __linux 122# if __linux
56# include <sys/sendfile.h> 123# include <sys/sendfile.h>
57# elif __freebsd 124# elif __FreeBSD__ || defined __APPLE__
58# include <sys/socket.h> 125# include <sys/socket.h>
59# include <sys/uio.h> 126# include <sys/uio.h>
60# elif __hpux 127# elif __hpux
61# include <sys/socket.h> 128# include <sys/socket.h>
62# elif __solaris /* not yet */ 129# elif __solaris
63# include <sys/sendfile.h> 130# include <sys/sendfile.h>
64# else 131# else
65# error sendfile support requested but not available 132# error sendfile support requested but not available
66# endif 133# endif
67#endif 134#endif
68 135
69/* number of seconds after which an idle threads exit */ 136#ifndef D_TYPE
70#define IDLE_TIMEOUT 10 137# define D_TYPE(de) 0
138#endif
139#ifndef D_INO
140# define D_INO(de) 0
141#endif
142#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name)
144#endif
71 145
72/* used for struct dirent, AIX doesn't provide it */ 146/* used for struct dirent, AIX doesn't provide it */
73#ifndef NAME_MAX 147#ifndef NAME_MAX
74# define NAME_MAX 4096 148# define NAME_MAX 4096
75#endif 149#endif
76 150
151/* used for readlink etc. */
152#ifndef PATH_MAX
153# define PATH_MAX 4096
154#endif
155
77/* buffer size for various temporary buffers */ 156/* buffer size for various temporary buffers */
78#define EIO_BUFSIZE 65536 157#define EIO_BUFSIZE 65536
79 158
80#define dBUF \ 159#define dBUF \
81 char *eio_buf; \ 160 char *eio_buf; \
82 X_LOCK (wrklock); \ 161 ETP_WORKER_LOCK (self); \
83 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \
84 X_UNLOCK (wrklock); \ 163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \
85 if (!eio_buf) \ 165 if (!eio_buf) \
86 return -1; 166 return -1;
87 167
88#define EIO_TICKS ((1000000 + 1023) >> 10) 168#define EIO_TICKS ((1000000 + 1023) >> 10)
89 169
170/*****************************************************************************/
171
172#if __GNUC__ >= 3
173# define expect(expr,value) __builtin_expect ((expr),(value))
174#else
175# define expect(expr,value) (expr)
176#endif
177
178#define expect_false(expr) expect ((expr) != 0, 0)
179#define expect_true(expr) expect ((expr) != 0, 1)
180
181/*****************************************************************************/
182
183#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX
185
186struct etp_worker;
187
188#define ETP_REQ eio_req
189#define ETP_DESTROY(req) eio_destroy (req)
190static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/
213
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215
216/* calculate time difference in ~1/EIO_TICKS of a second */
217static int tvdiff (struct timeval *tv1, struct timeval *tv2)
218{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221}
222
223static unsigned int started, idle, wanted = 4;
224
90static void (*want_poll_cb) (void); 225static void (*want_poll_cb) (void);
91static void (*done_poll_cb) (void); 226static void (*done_poll_cb) (void);
92 227
93static unsigned int max_poll_time = 0; 228static unsigned int max_poll_time; /* reslock */
94static unsigned int max_poll_reqs = 0; 229static unsigned int max_poll_reqs; /* reslock */
95 230
96/* calculcate time difference in ~1/EIO_TICKS of a second */ 231static volatile unsigned int nreqs; /* reqlock */
97static int tvdiff (struct timeval *tv1, struct timeval *tv2) 232static volatile unsigned int nready; /* reqlock */
98{ 233static volatile unsigned int npending; /* reqlock */
99 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
100 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
101}
102 236
103static unsigned int started, idle, wanted; 237static xmutex_t wrklock;
238static xmutex_t reslock;
239static xmutex_t reqlock;
240static xcond_t reqwait;
241
242#if !HAVE_PREADWRITE
243/*
244 * make our pread/pwrite emulation safe against themselves, but not against
245 * normal read/write by using a mutex. slows down execution a lot,
246 * but that's your problem, not mine.
247 */
248static xmutex_t preadwritelock = X_MUTEX_INIT;
249#endif
250
251typedef struct etp_worker
252{
253 /* locked by wrklock */
254 struct etp_worker *prev, *next;
255
256 xthread_t tid;
257
258 /* locked by reslock, reqlock or wrklock */
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON
262} etp_worker;
263
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
265
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
104 268
105/* worker threads management */ 269/* worker threads management */
106static mutex_t wrklock = X_MUTEX_INIT;
107 270
108typedef struct worker {
109 /* locked by wrklock */
110 struct worker *prev, *next;
111
112 thread_t tid;
113
114 /* locked by reslock, reqlock or wrklock */
115 eio_req *req; /* currently processed request */
116 void *dbuf;
117 DIR *dirp;
118} worker;
119
120static worker wrk_first = { &wrk_first, &wrk_first, 0 };
121
122static void worker_clear (worker *wrk) 271static void etp_worker_clear (etp_worker *wrk)
123{ 272{
124 if (wrk->dirp) 273 ETP_WORKER_CLEAR (wrk);
125 {
126 closedir (wrk->dirp);
127 wrk->dirp = 0;
128 }
129
130 if (wrk->dbuf)
131 {
132 free (wrk->dbuf);
133 wrk->dbuf = 0;
134 }
135} 274}
136 275
137static void worker_free (worker *wrk) 276static void etp_worker_free (etp_worker *wrk)
138{ 277{
139 wrk->next->prev = wrk->prev; 278 wrk->next->prev = wrk->prev;
140 wrk->prev->next = wrk->next; 279 wrk->prev->next = wrk->next;
141 280
142 free (wrk); 281 free (wrk);
143} 282}
144 283
145static volatile unsigned int nreqs, nready, npending;
146static volatile unsigned int max_idle = 4;
147
148static mutex_t reslock = X_MUTEX_INIT;
149static mutex_t reqlock = X_MUTEX_INIT;
150static cond_t reqwait = X_COND_INIT;
151
152unsigned int eio_nreqs (void) 284static unsigned int etp_nreqs (void)
153{ 285{
286 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
154 return nreqs; 290 return retval;
155} 291}
156 292
157unsigned int eio_nready (void) 293static unsigned int etp_nready (void)
158{ 294{
159 unsigned int retval; 295 unsigned int retval;
160 296
161 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
162 retval = nready; 298 retval = nready;
163 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
164 300
165 return retval; 301 return retval;
166} 302}
167 303
168unsigned int eio_npending (void) 304static unsigned int etp_npending (void)
169{ 305{
170 unsigned int retval; 306 unsigned int retval;
171 307
172 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
173 retval = npending; 309 retval = npending;
174 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
175 311
176 return retval; 312 return retval;
177} 313}
178 314
179unsigned int eio_nthreads (void) 315static unsigned int etp_nthreads (void)
180{ 316{
181 unsigned int retval; 317 unsigned int retval;
182 318
183 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
184 retval = started; 320 retval = started;
… …
191 * a somewhat faster data structure might be nice, but 327 * a somewhat faster data structure might be nice, but
192 * with 8 priorities this actually needs <20 insns 328 * with 8 priorities this actually needs <20 insns
193 * per shift, the most expensive operation. 329 * per shift, the most expensive operation.
194 */ 330 */
195typedef struct { 331typedef struct {
196 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ 332 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
197 int size; 333 int size;
198} reqq; 334} etp_reqq;
199 335
200static reqq req_queue; 336static etp_reqq req_queue;
201static reqq res_queue; 337static etp_reqq res_queue;
202 338
203static int reqq_push (reqq *q, eio_req *req) 339static int reqq_push (etp_reqq *q, ETP_REQ *req)
204{ 340{
205 int pri = req->pri; 341 int pri = req->pri;
206 req->next = 0; 342 req->next = 0;
207 343
208 if (q->qe[pri]) 344 if (q->qe[pri])
… …
214 q->qe[pri] = q->qs[pri] = req; 350 q->qe[pri] = q->qs[pri] = req;
215 351
216 return q->size++; 352 return q->size++;
217} 353}
218 354
219static eio_req *reqq_shift (reqq *q) 355static ETP_REQ *reqq_shift (etp_reqq *q)
220{ 356{
221 int pri; 357 int pri;
222 358
223 if (!q->size) 359 if (!q->size)
224 return 0; 360 return 0;
225 361
226 --q->size; 362 --q->size;
227 363
228 for (pri = EIO_NUM_PRI; pri--; ) 364 for (pri = ETP_NUM_PRI; pri--; )
229 { 365 {
230 eio_req *req = q->qs[pri]; 366 eio_req *req = q->qs[pri];
231 367
232 if (req) 368 if (req)
233 { 369 {
… …
239 } 375 }
240 376
241 abort (); 377 abort ();
242} 378}
243 379
244static void grp_try_feed (eio_req *grp) 380static void etp_thread_init (void)
245{ 381{
246 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 382 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait);
386}
387
388static void etp_atfork_prepare (void)
389{
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397
398static void etp_atfork_parent (void)
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
247 { 419 {
248 int old_len = grp->size; 420 etp_worker *wrk = wrk_first.next;
249 421
250 EIO_FEED (grp); 422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
251 424
252 /* stop if no progress has been made */ 425 etp_worker_clear (wrk);
253 if (old_len == grp->size) 426 etp_worker_free (wrk);
254 {
255 grp->feed = 0;
256 break;
257 }
258 } 427 }
259}
260 428
261static int eio_finish (eio_req *req); 429 started = 0;
430 idle = 0;
431 nreqs = 0;
432 nready = 0;
433 npending = 0;
262 434
263static int grp_dec (eio_req *grp) 435 etp_thread_init ();
264{ 436}
265 --grp->size;
266 437
267 /* call feeder, if applicable */ 438static void
268 grp_try_feed (grp); 439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
269 444
270 /* finish, if done */ 445static int
271 if (!grp->size && grp->int1) 446etp_init (void (*want_poll)(void), void (*done_poll)(void))
272 return eio_finish (grp); 447{
273 else 448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll;
453 done_poll_cb = done_poll;
454
274 return 0; 455 return 0;
275} 456}
276 457
277void eio_destroy (eio_req *req)
278{
279 if ((req)->flags & EIO_FLAG_PTR2_FREE)
280 free (req->ptr2);
281
282 EIO_DESTROY (req);
283}
284
285static int eio_finish (eio_req *req)
286{
287 int res = EIO_FINISH (req);
288
289 if (req->grp)
290 {
291 int res2;
292 eio_req *grp = req->grp;
293
294 /* unlink request */
295 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
296 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
297
298 if (grp->grp_first == req)
299 grp->grp_first = req->grp_next;
300
301 res2 = grp_dec (grp);
302
303 if (!res && res2)
304 res = res2;
305 }
306
307 eio_destroy (req);
308
309 return res;
310}
311
312void eio_grp_cancel (eio_req *grp)
313{
314 for (grp = grp->grp_first; grp; grp = grp->grp_next)
315 eio_cancel (grp);
316}
317
318void eio_cancel (eio_req *req)
319{
320 req->flags |= EIO_FLAG_CANCELLED;
321
322 eio_grp_cancel (req);
323}
324
325X_THREAD_PROC (eio_proc); 458X_THREAD_PROC (etp_proc);
326 459
327static void start_thread (void) 460static void etp_start_thread (void)
328{ 461{
329 worker *wrk = calloc (1, sizeof (worker)); 462 etp_worker *wrk = calloc (1, sizeof (etp_worker));
330 463
331 if (!wrk) 464 /*TODO*/
332 croak ("unable to allocate worker thread data"); 465 assert (("unable to allocate worker thread data", wrk));
333 466
334 X_LOCK (wrklock); 467 X_LOCK (wrklock);
335 468
336 if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) 469 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
337 { 470 {
338 wrk->prev = &wrk_first; 471 wrk->prev = &wrk_first;
339 wrk->next = wrk_first.next; 472 wrk->next = wrk_first.next;
340 wrk_first.next->prev = wrk; 473 wrk_first.next->prev = wrk;
341 wrk_first.next = wrk; 474 wrk_first.next = wrk;
… …
345 free (wrk); 478 free (wrk);
346 479
347 X_UNLOCK (wrklock); 480 X_UNLOCK (wrklock);
348} 481}
349 482
350static void maybe_start_thread (void) 483static void etp_maybe_start_thread (void)
351{ 484{
352 if (eio_nthreads () >= wanted) 485 if (expect_true (etp_nthreads () >= wanted))
353 return; 486 return;
354 487
355 /* todo: maybe use idle here, but might be less exact */ 488 /* todo: maybe use idle here, but might be less exact */
356 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) 489 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
357 return; 490 return;
358 491
359 start_thread (); 492 etp_start_thread ();
360} 493}
361 494
362void eio_submit (eio_req *req)
363{
364 ++nreqs;
365
366 X_LOCK (reqlock);
367 ++nready;
368 reqq_push (&req_queue, req);
369 X_COND_SIGNAL (reqwait);
370 X_UNLOCK (reqlock);
371
372 maybe_start_thread ();
373}
374
375static void end_thread (void) 495static void etp_end_thread (void)
376{ 496{
377 eio_req *req = calloc (1, sizeof (eio_req)); 497 eio_req *req = calloc (1, sizeof (eio_req));
378 498
379 req->type = EIO_QUIT; 499 req->type = -1;
380 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; 500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
381 501
382 X_LOCK (reqlock); 502 X_LOCK (reqlock);
383 reqq_push (&req_queue, req); 503 reqq_push (&req_queue, req);
384 X_COND_SIGNAL (reqwait); 504 X_COND_SIGNAL (reqwait);
385 X_UNLOCK (reqlock); 505 X_UNLOCK (reqlock);
… …
387 X_LOCK (wrklock); 507 X_LOCK (wrklock);
388 --started; 508 --started;
389 X_UNLOCK (wrklock); 509 X_UNLOCK (wrklock);
390} 510}
391 511
392void eio_set_max_poll_time (double nseconds)
393{
394 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
395 max_poll_time = nseconds;
396 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
397}
398
399void eio_set_max_poll_reqs (unsigned int maxreqs)
400{
401 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
402 max_poll_reqs = maxreqs;
403 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
404}
405
406void eio_set_max_idle (unsigned int nthreads)
407{
408 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
409 max_idle = nthreads <= 0 ? 1 : nthreads;
410 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
411}
412
413void eio_set_min_parallel (unsigned int nthreads)
414{
415 if (wanted < nthreads)
416 wanted = nthreads;
417}
418
419void eio_set_max_parallel (unsigned int nthreads)
420{
421 if (wanted > nthreads)
422 wanted = nthreads;
423
424 while (started > wanted)
425 end_thread ();
426}
427
428int eio_poll (void) 512static int etp_poll (void)
429{ 513{
430 int maxreqs = max_poll_reqs; 514 unsigned int maxreqs;
515 unsigned int maxtime;
431 struct timeval tv_start, tv_now; 516 struct timeval tv_start, tv_now;
432 eio_req *req;
433 517
518 X_LOCK (reslock);
519 maxreqs = max_poll_reqs;
520 maxtime = max_poll_time;
521 X_UNLOCK (reslock);
522
434 if (max_poll_time) 523 if (maxtime)
435 gettimeofday (&tv_start, 0); 524 gettimeofday (&tv_start, 0);
436 525
437 for (;;) 526 for (;;)
438 { 527 {
528 ETP_REQ *req;
529
439 maybe_start_thread (); 530 etp_maybe_start_thread ();
440 531
441 X_LOCK (reslock); 532 X_LOCK (reslock);
442 req = reqq_shift (&res_queue); 533 req = reqq_shift (&res_queue);
443 534
444 if (req) 535 if (req)
… …
452 X_UNLOCK (reslock); 543 X_UNLOCK (reslock);
453 544
454 if (!req) 545 if (!req)
455 return 0; 546 return 0;
456 547
548 X_LOCK (reqlock);
457 --nreqs; 549 --nreqs;
550 X_UNLOCK (reqlock);
458 551
459 if (req->type == EIO_GROUP && req->size) 552 if (expect_false (req->type == EIO_GROUP && req->size))
460 { 553 {
461 req->int1 = 1; /* mark request as delayed */ 554 req->int1 = 1; /* mark request as delayed */
462 continue; 555 continue;
463 } 556 }
464 else 557 else
465 { 558 {
466 int res = eio_finish (req); 559 int res = ETP_FINISH (req);
467 if (res) 560 if (expect_false (res))
468 return res; 561 return res;
469 } 562 }
470 563
471 if (maxreqs && !--maxreqs) 564 if (expect_false (maxreqs && !--maxreqs))
472 break; 565 break;
473 566
474 if (max_poll_time) 567 if (maxtime)
475 { 568 {
476 gettimeofday (&tv_now, 0); 569 gettimeofday (&tv_now, 0);
477 570
478 if (tvdiff (&tv_start, &tv_now) >= max_poll_time) 571 if (tvdiff (&tv_start, &tv_now) >= maxtime)
479 break; 572 break;
480 } 573 }
481 } 574 }
482 575
483 errno = EAGAIN; 576 errno = EAGAIN;
484 return -1; 577 return -1;
485} 578}
486 579
580static void etp_cancel (ETP_REQ *req)
581{
582 X_LOCK (wrklock);
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585
586 eio_grp_cancel (req);
587}
588
589static void etp_submit (ETP_REQ *req)
590{
591 req->pri -= ETP_PRI_MIN;
592
593 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
594 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
595
596 if (expect_false (req->type == EIO_GROUP))
597 {
598 /* I hope this is worth it :/ */
599 X_LOCK (reqlock);
600 ++nreqs;
601 X_UNLOCK (reqlock);
602
603 X_LOCK (reslock);
604
605 ++npending;
606
607 if (!reqq_push (&res_queue, req) && want_poll_cb)
608 want_poll_cb ();
609
610 X_UNLOCK (reslock);
611 }
612 else
613 {
614 X_LOCK (reqlock);
615 ++nreqs;
616 ++nready;
617 reqq_push (&req_queue, req);
618 X_COND_SIGNAL (reqwait);
619 X_UNLOCK (reqlock);
620
621 etp_maybe_start_thread ();
622 }
623}
624
625static void etp_set_max_poll_time (double nseconds)
626{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630}
631
632static void etp_set_max_poll_reqs (unsigned int maxreqs)
633{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637}
638
639static void etp_set_max_idle (unsigned int nthreads)
640{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644}
645
646static void etp_set_idle_timeout (unsigned int seconds)
647{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651}
652
653static void etp_set_min_parallel (unsigned int nthreads)
654{
655 if (wanted < nthreads)
656 wanted = nthreads;
657}
658
659static void etp_set_max_parallel (unsigned int nthreads)
660{
661 if (wanted > nthreads)
662 wanted = nthreads;
663
664 while (started > wanted)
665 etp_end_thread ();
666}
667
668/*****************************************************************************/
669
670static void grp_try_feed (eio_req *grp)
671{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 {
674 grp->flags &= ~EIO_FLAG_GROUPADD;
675
676 EIO_FEED (grp);
677
678 /* stop if no progress has been made */
679 if (!(grp->flags & EIO_FLAG_GROUPADD))
680 {
681 grp->feed = 0;
682 break;
683 }
684 }
685}
686
687static int grp_dec (eio_req *grp)
688{
689 --grp->size;
690
691 /* call feeder, if applicable */
692 grp_try_feed (grp);
693
694 /* finish, if done */
695 if (!grp->size && grp->int1)
696 return eio_finish (grp);
697 else
698 return 0;
699}
700
701void eio_destroy (eio_req *req)
702{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705
706 EIO_DESTROY (req);
707}
708
709static int eio_finish (eio_req *req)
710{
711 int res = EIO_FINISH (req);
712
713 if (req->grp)
714 {
715 int res2;
716 eio_req *grp = req->grp;
717
718 /* unlink request */
719 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
720 if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
721
722 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next;
724
725 res2 = grp_dec (grp);
726
727 if (!res && res2)
728 res = res2;
729 }
730
731 eio_destroy (req);
732
733 return res;
734}
735
736void eio_grp_cancel (eio_req *grp)
737{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp);
740}
741
742void eio_cancel (eio_req *req)
743{
744 etp_cancel (req);
745}
746
747void eio_submit (eio_req *req)
748{
749 etp_submit (req);
750}
751
752unsigned int eio_nreqs (void)
753{
754 return etp_nreqs ();
755}
756
757unsigned int eio_nready (void)
758{
759 return etp_nready ();
760}
761
762unsigned int eio_npending (void)
763{
764 return etp_npending ();
765}
766
767unsigned int eio_nthreads (void)
768{
769 return etp_nthreads ();
770}
771
772void eio_set_max_poll_time (double nseconds)
773{
774 etp_set_max_poll_time (nseconds);
775}
776
777void eio_set_max_poll_reqs (unsigned int maxreqs)
778{
779 etp_set_max_poll_reqs (maxreqs);
780}
781
782void eio_set_max_idle (unsigned int nthreads)
783{
784 etp_set_max_idle (nthreads);
785}
786
787void eio_set_idle_timeout (unsigned int seconds)
788{
789 etp_set_idle_timeout (seconds);
790}
791
792void eio_set_min_parallel (unsigned int nthreads)
793{
794 etp_set_min_parallel (nthreads);
795}
796
797void eio_set_max_parallel (unsigned int nthreads)
798{
799 etp_set_max_parallel (nthreads);
800}
801
802int eio_poll (void)
803{
804 return etp_poll ();
805}
806
487/*****************************************************************************/ 807/*****************************************************************************/
488/* work around various missing functions */ 808/* work around various missing functions */
489 809
490#if !HAVE_PREADWRITE 810#if !HAVE_PREADWRITE
811# undef pread
812# undef pwrite
491# define pread aio_pread 813# define pread eio__pread
492# define pwrite aio_pwrite 814# define pwrite eio__pwrite
493 815
494/* 816static ssize_t
495 * make our pread/pwrite safe against themselves, but not against 817eio__pread (int fd, void *buf, size_t count, off_t offset)
496 * normal read/write by using a mutex. slows down execution a lot,
497 * but that's your problem, not mine.
498 */
499static mutex_t preadwritelock = X_MUTEX_INIT;
500
501static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
502{ 818{
503 ssize_t res; 819 ssize_t res;
504 off_t ooffset; 820 off_t ooffset;
505 821
506 X_LOCK (preadwritelock); 822 X_LOCK (preadwritelock);
… …
511 X_UNLOCK (preadwritelock); 827 X_UNLOCK (preadwritelock);
512 828
513 return res; 829 return res;
514} 830}
515 831
516static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) 832static ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset)
517{ 834{
518 ssize_t res; 835 ssize_t res;
519 off_t ooffset; 836 off_t ooffset;
520 837
521 X_LOCK (preadwritelock); 838 X_LOCK (preadwritelock);
522 ooffset = lseek (fd, 0, SEEK_CUR); 839 ooffset = lseek (fd, 0, SEEK_CUR);
523 lseek (fd, offset, SEEK_SET); 840 lseek (fd, offset, SEEK_SET);
524 res = write (fd, buf, count); 841 res = write (fd, buf, count);
525 lseek (fd, offset, SEEK_SET); 842 lseek (fd, ooffset, SEEK_SET);
526 X_UNLOCK (preadwritelock); 843 X_UNLOCK (preadwritelock);
527 844
528 return res; 845 return res;
529} 846}
530#endif 847#endif
531 848
532#ifndef HAVE_FUTIMES 849#ifndef HAVE_UTIMES
533 850
851# undef utimes
534# define utimes(path,times) aio_utimes (path, times) 852# define utimes(path,times) eio__utimes (path, times)
535# define futimes(fd,times) aio_futimes (fd, times)
536 853
854static int
537static int aio_utimes (const char *filename, const struct timeval times[2]) 855eio__utimes (const char *filename, const struct timeval times[2])
538{ 856{
539 if (times) 857 if (times)
540 { 858 {
541 struct utimbuf buf; 859 struct utimbuf buf;
542 860
… …
547 } 865 }
548 else 866 else
549 return utime (filename, 0); 867 return utime (filename, 0);
550} 868}
551 869
870#endif
871
872#ifndef HAVE_FUTIMES
873
874# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times)
876
552static int aio_futimes (int fd, const struct timeval tv[2]) 877static int eio__futimes (int fd, const struct timeval tv[2])
553{ 878{
554 errno = ENOSYS; 879 errno = ENOSYS;
555 return -1; 880 return -1;
556} 881}
557 882
558#endif 883#endif
559 884
560#if !HAVE_FDATASYNC 885#if !HAVE_FDATASYNC
886# undef fdatasync
561# define fdatasync fsync 887# define fdatasync(fd) fsync (fd)
562#endif 888#endif
889
890/* sync_file_range always needs emulation */
891int
892eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
893{
894#if HAVE_SYNC_FILE_RANGE
895 int res;
896
897 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
898 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
899 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
900 {
901 flags = 0
902 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
903 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
904 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
905 }
906
907 res = sync_file_range (fd, offset, nbytes, flags);
908
909 if (!res || errno != ENOSYS)
910 return res;
911#endif
912
913 /* even though we could play tricks with the flags, it's better to always
914 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd);
916}
563 917
564#if !HAVE_READAHEAD 918#if !HAVE_READAHEAD
919# undef readahead
565# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) 920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 921
922static ssize_t
567static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) 923eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
568{ 924{
569 size_t todo = count; 925 size_t todo = count;
570 dBUF; 926 dBUF;
571 927
572 while (todo > 0) 928 while (todo > 0)
573 { 929 {
574 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 930 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
575 931
576 pread (fd, aio_buf, len, offset); 932 pread (fd, eio_buf, len, offset);
577 offset += len; 933 offset += len;
578 todo -= len; 934 todo -= len;
579 } 935 }
580 936
581 errno = 0; 937 errno = 0;
582 return count; 938 return count;
583} 939}
584 940
585#endif 941#endif
586 942
587#if !HAVE_READDIR_R
588# define readdir_r aio_readdir_r
589
590static mutex_t readdirlock = X_MUTEX_INIT;
591
592static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res)
593{
594 X_DIRENT *e;
595 int errorno;
596
597 X_LOCK (readdirlock);
598
599 e = readdir (dirp);
600 errorno = errno;
601
602 if (e)
603 {
604 *res = ent;
605 strcpy (ent->d_name, e->d_name);
606 }
607 else
608 *res = 0;
609
610 X_UNLOCK (readdirlock);
611
612 errno = errorno;
613 return e ? 0 : -1;
614}
615#endif
616
617/* sendfile always needs emulation */ 943/* sendfile always needs emulation */
944static ssize_t
618static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) 945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
619{ 946{
947 ssize_t written = 0;
620 ssize_t res; 948 ssize_t res;
621 949
622 if (!count) 950 if (!count)
623 return 0; 951 return 0;
624 952
953 for (;;)
954 {
625#if HAVE_SENDFILE 955#if HAVE_SENDFILE
626# if __linux 956# if __linux
957 off_t soffset = offset;
627 res = sendfile (ofd, ifd, &offset, count); 958 res = sendfile (ofd, ifd, &soffset, count);
628 959
629# elif __freebsd 960# elif __FreeBSD__
630 /* 961 /*
631 * Of course, the freebsd sendfile is a dire hack with no thoughts 962 * Of course, the freebsd sendfile is a dire hack with no thoughts
632 * wasted on making it similar to other I/O functions. 963 * wasted on making it similar to other I/O functions.
633 */ 964 */
634 {
635 off_t sbytes; 965 off_t sbytes;
636 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 966 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
637 967
638 if (res < 0 && sbytes) 968 #if 0 /* according to the manpage, this is correct, but broken behaviour */
639 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 969 /* freebsd' sendfile will return 0 on success */
970 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
971 /* not on e.g. EIO or EPIPE - sounds broken */
972 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
640 res = sbytes; 973 res = sbytes;
641 } 974 #endif
975
976 /* according to source inspection, this is correct, and useful behaviour */
977 if (sbytes)
978 res = sbytes;
979
980# elif defined (__APPLE__)
981 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983
984 /* according to the manpage, sbytes is always valid */
985 if (sbytes)
986 res = sbytes;
642 987
643# elif __hpux 988# elif __hpux
644 res = sendfile (ofd, ifd, offset, count, 0, 0); 989 res = sendfile (ofd, ifd, offset, count, 0, 0);
645 990
646# elif __solaris 991# elif __solaris
647 {
648 struct sendfilevec vec; 992 struct sendfilevec vec;
649 size_t sbytes; 993 size_t sbytes;
650 994
651 vec.sfv_fd = ifd; 995 vec.sfv_fd = ifd;
652 vec.sfv_flag = 0; 996 vec.sfv_flag = 0;
653 vec.sfv_off = offset; 997 vec.sfv_off = offset;
654 vec.sfv_len = count; 998 vec.sfv_len = count;
655 999
656 res = sendfilev (ofd, &vec, 1, &sbytes); 1000 res = sendfilev (ofd, &vec, 1, &sbytes);
657 1001
658 if (res < 0 && sbytes) 1002 if (res < 0 && sbytes)
659 res = sbytes; 1003 res = sbytes;
660 }
661 1004
662# endif 1005# endif
1006
1007#elif defined (_WIN32)
1008 /* does not work, just for documentation of what would need to be done */
1009 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1010 /* libeio guarantees that the file offset does not change, and windows */
1011 /* has no way to get an independent handle to the same file description */
1012 HANDLE h = TO_SOCKET (ifd);
1013 SetFilePointer (h, offset, 0, FILE_BEGIN);
1014 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1015
663#else 1016#else
664 res = -1; 1017 res = -1;
665 errno = ENOSYS; 1018 errno = ENOSYS;
666#endif 1019#endif
667 1020
1021 /* we assume sendfile can copy at least 128mb in one go */
1022 if (res <= 128 * 1024 * 1024)
1023 {
1024 if (res > 0)
1025 written += res;
1026
1027 if (written)
1028 return written;
1029
1030 break;
1031 }
1032 else
1033 {
1034 /* if we requested more, then probably the kernel was lazy */
1035 written += res;
1036 offset += res;
1037 count -= res;
1038
1039 if (!count)
1040 return written;
1041 }
1042 }
1043
668 if (res < 0 1044 if (res < 0
669 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1045 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1046 /* BSDs */
1047#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1048 || errno == ENOTSUP
1049#endif
1050 || errno == EOPNOTSUPP /* BSDs */
670#if __solaris 1051#if __solaris
671 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1052 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
672#endif 1053#endif
673 ) 1054 )
674 ) 1055 )
… …
705 } 1086 }
706 1087
707 return res; 1088 return res;
708} 1089}
709 1090
1091static signed char
1092eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1093{
1094 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1095 : a->inode < b->inode ? -1
1096 : a->inode > b->inode ? 1
1097 : 0;
1098}
1099
1100#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1101
1102#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1103#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1104
1105static void
1106eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1107{
1108 unsigned char bits [9 + sizeof (ino_t) * 8];
1109 unsigned char *bit = bits;
1110
1111 assert (CHAR_BIT == 8);
1112 assert (sizeof (eio_dirent) * 8 < 256);
1113 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1114 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1115
1116 if (size <= EIO_SORT_FAST)
1117 return;
1118
1119 /* first prepare an array of bits to test in our radix sort */
1120 /* try to take endianness into account, as well as differences in ino_t sizes */
1121 /* inode_bits must contain all inodes ORed together */
1122 /* which is used to skip bits that are 0 everywhere, which is very common */
1123 {
1124 ino_t endianness;
1125 int i, j;
1126
1127 /* we store the byte offset of byte n into byte n of "endianness" */
1128 for (i = 0; i < sizeof (ino_t); ++i)
1129 ((unsigned char *)&endianness)[i] = i;
1130
1131 *bit++ = 0;
1132
1133 for (i = 0; i < sizeof (ino_t); ++i)
1134 {
1135 /* shifting off the byte offsets out of "endianness" */
1136 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1137 endianness >>= 8;
1138
1139 for (j = 0; j < 8; ++j)
1140 if (inode_bits & (((ino_t)1) << (i * 8 + j)))
1141 *bit++ = offs + j;
1142 }
1143
1144 for (j = 0; j < 8; ++j)
1145 if (score_bits & (1 << j))
1146 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1147 }
1148
1149 /* now actually do the sorting (a variant of MSD radix sort) */
1150 {
1151 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base;
1152 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end;
1153 unsigned char *bit_stk [9 + sizeof (ino_t) * 8];
1154 int stk_idx = 0;
1155
1156 base_stk [stk_idx] = dents;
1157 end_stk [stk_idx] = dents + size;
1158 bit_stk [stk_idx] = bit - 1;
1159
1160 do
1161 {
1162 base = base_stk [stk_idx];
1163 end = end_stk [stk_idx];
1164 bit = bit_stk [stk_idx];
1165
1166 for (;;)
1167 {
1168 unsigned char O = *bit >> 3;
1169 unsigned char M = 1 << (*bit & 7);
1170
1171 eio_dirent *a = base;
1172 eio_dirent *b = end;
1173
1174 if (b - a < EIO_SORT_CUTOFF)
1175 break;
1176
1177 /* now bit-partition the array on the bit */
1178 /* this ugly asymmetric loop seems to perform much better than typical */
1179 /* partition algos found in the literature */
1180 do
1181 if (!(((unsigned char *)a)[O] & M))
1182 ++a;
1183 else if (!(((unsigned char *)--b)[O] & M))
1184 {
1185 eio_dirent tmp = *a; *a = *b; *b = tmp;
1186 ++a;
1187 }
1188 while (b > a);
1189
1190 /* next bit, or stop, if no bits left in this path */
1191 if (!*--bit)
1192 break;
1193
1194 base_stk [stk_idx] = a;
1195 end_stk [stk_idx] = end;
1196 bit_stk [stk_idx] = bit;
1197 ++stk_idx;
1198
1199 end = a;
1200 }
1201 }
1202 while (stk_idx--);
1203 }
1204}
1205
1206static void
1207eio_dent_insertion_sort (eio_dirent *dents, int size)
1208{
1209 /* first move the smallest element to the front, to act as a sentinel */
1210 {
1211 int i;
1212 eio_dirent *min = dents;
1213
1214 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1215 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1216 if (EIO_DENT_CMP (dents [i], <, *min))
1217 min = &dents [i];
1218
1219 /* swap elements 0 and j (minimum) */
1220 {
1221 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1222 }
1223 }
1224
1225 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1226 {
1227 eio_dirent *i, *j;
1228
1229 for (i = dents + 1; i < dents + size; ++i)
1230 {
1231 eio_dirent value = *i;
1232
1233 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1234 j [1] = j [0];
1235
1236 j [1] = value;
1237 }
1238 }
1239}
1240
1241static void
1242eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
1243{
1244 if (size <= 1)
1245 return; /* our insertion sort relies on size > 0 */
1246
1247 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1248 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1249 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1250
1251 /* use an insertion sort at the end, or for small arrays, */
1252 /* as insertion sort is more efficient for small partitions */
1253 eio_dent_insertion_sort (dents, size);
1254}
1255
710/* read a full directory */ 1256/* read a full directory */
1257static void
711static void scandir_ (eio_req *req, worker *self) 1258eio__scandir (eio_req *req, etp_worker *self)
712{ 1259{
713 DIR *dirp; 1260 DIR *dirp;
714 union
715 {
716 EIO_STRUCT_DIRENT d;
717 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
718 } *u;
719 EIO_STRUCT_DIRENT *entp; 1261 EIO_STRUCT_DIRENT *entp;
720 char *name, *names; 1262 char *name, *names;
721 int memlen = 4096; 1263 int namesalloc = 4096;
722 int memofs = 0; 1264 int namesoffs = 0;
1265 int flags = req->int1;
1266 eio_dirent *dents = 0;
1267 int dentalloc = 128;
723 int res = 0; 1268 int dentoffs = 0;
1269 ino_t inode_bits = 0;
1270
1271 req->result = -1;
1272
1273 if (!(flags & EIO_READDIR_DENTS))
1274 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
724 1275
725 X_LOCK (wrklock); 1276 X_LOCK (wrklock);
1277 /* the corresponding closedir is in ETP_WORKER_CLEAR */
726 self->dirp = dirp = opendir (req->ptr1); 1278 self->dirp = dirp = opendir (req->ptr1);
727 self->dbuf = u = malloc (sizeof (*u)); 1279
728 req->flags |= EIO_FLAG_PTR2_FREE; 1280 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1281 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
729 req->ptr2 = names = malloc (memlen); 1282 req->ptr2 = names = malloc (namesalloc);
730 X_UNLOCK (wrklock); 1283 X_UNLOCK (wrklock);
731 1284
732 if (dirp && u && names) 1285 if (dirp && names && (!flags || dents))
733 for (;;) 1286 for (;;)
734 { 1287 {
735 errno = 0; 1288 errno = 0;
736 readdir_r (dirp, &u->d, &entp); 1289 entp = readdir (dirp);
737 1290
738 if (!entp) 1291 if (!entp)
1292 {
1293 if (errno)
1294 break;
1295
1296 /* sort etc. */
1297 req->int1 = flags;
1298 req->result = dentoffs;
1299
1300 if (flags & EIO_READDIR_STAT_ORDER)
1301 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1302 else if (flags & EIO_READDIR_DIRS_FIRST)
1303 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1304 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1305 else
1306 {
1307 /* in this case, all is known, and we just put dirs first and sort them */
1308 eio_dirent *oth = dents + dentoffs;
1309 eio_dirent *dir = dents;
1310
1311 /* now partition dirs to the front, and non-dirs to the back */
1312 /* by walking from both sides and swapping if necessary */
1313 while (oth > dir)
1314 {
1315 if (dir->type == EIO_DT_DIR)
1316 ++dir;
1317 else if ((--oth)->type == EIO_DT_DIR)
1318 {
1319 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1320
1321 ++dir;
1322 }
1323 }
1324
1325 /* now sort the dirs only (dirs all have the same score) */
1326 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1327 }
1328
739 break; 1329 break;
1330 }
740 1331
1332 /* now add the entry to our list(s) */
741 name = entp->d_name; 1333 name = entp->d_name;
742 1334
1335 /* skip . and .. entries */
743 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1336 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
744 { 1337 {
745 int len = strlen (name) + 1; 1338 int len = D_NAMLEN (entp) + 1;
746 1339
747 res++; 1340 while (expect_false (namesoffs + len > namesalloc))
748
749 while (memofs + len > memlen)
750 { 1341 {
751 memlen *= 2; 1342 namesalloc *= 2;
752 X_LOCK (wrklock); 1343 X_LOCK (wrklock);
753 req->ptr2 = names = realloc (names, memlen); 1344 req->ptr2 = names = realloc (names, namesalloc);
754 X_UNLOCK (wrklock); 1345 X_UNLOCK (wrklock);
755 1346
756 if (!names) 1347 if (!names)
757 break; 1348 break;
758 } 1349 }
759 1350
760 memcpy (names + memofs, name, len); 1351 memcpy (names + namesoffs, name, len);
1352
1353 if (dents)
1354 {
1355 struct eio_dirent *ent;
1356
1357 if (expect_false (dentoffs == dentalloc))
1358 {
1359 dentalloc *= 2;
1360 X_LOCK (wrklock);
1361 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1362 X_UNLOCK (wrklock);
1363
1364 if (!dents)
1365 break;
1366 }
1367
1368 ent = dents + dentoffs;
1369
1370 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1371 ent->namelen = len - 1;
1372 ent->inode = D_INO (entp);
1373
1374 inode_bits |= ent->inode;
1375
1376 switch (D_TYPE (entp))
1377 {
1378 default:
1379 ent->type = EIO_DT_UNKNOWN;
1380 flags |= EIO_READDIR_FOUND_UNKNOWN;
1381 break;
1382
1383 #ifdef DT_FIFO
1384 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1385 #endif
1386 #ifdef DT_CHR
1387 case DT_CHR: ent->type = EIO_DT_CHR; break;
1388 #endif
1389 #ifdef DT_MPC
1390 case DT_MPC: ent->type = EIO_DT_MPC; break;
1391 #endif
1392 #ifdef DT_DIR
1393 case DT_DIR: ent->type = EIO_DT_DIR; break;
1394 #endif
1395 #ifdef DT_NAM
1396 case DT_NAM: ent->type = EIO_DT_NAM; break;
1397 #endif
1398 #ifdef DT_BLK
1399 case DT_BLK: ent->type = EIO_DT_BLK; break;
1400 #endif
1401 #ifdef DT_MPB
1402 case DT_MPB: ent->type = EIO_DT_MPB; break;
1403 #endif
1404 #ifdef DT_REG
1405 case DT_REG: ent->type = EIO_DT_REG; break;
1406 #endif
1407 #ifdef DT_NWK
1408 case DT_NWK: ent->type = EIO_DT_NWK; break;
1409 #endif
1410 #ifdef DT_CMP
1411 case DT_CMP: ent->type = EIO_DT_CMP; break;
1412 #endif
1413 #ifdef DT_LNK
1414 case DT_LNK: ent->type = EIO_DT_LNK; break;
1415 #endif
1416 #ifdef DT_SOCK
1417 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1418 #endif
1419 #ifdef DT_DOOR
1420 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1421 #endif
1422 #ifdef DT_WHT
1423 case DT_WHT: ent->type = EIO_DT_WHT; break;
1424 #endif
1425 }
1426
1427 ent->score = 7;
1428
1429 if (flags & EIO_READDIR_DIRS_FIRST)
1430 {
1431 if (ent->type == EIO_DT_UNKNOWN)
1432 {
1433 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1434 ent->score = 1;
1435 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1436 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1437 }
1438 else if (ent->type == EIO_DT_DIR)
1439 ent->score = 0;
1440 }
1441 }
1442
761 memofs += len; 1443 namesoffs += len;
1444 ++dentoffs;
1445 }
1446
1447 if (EIO_CANCELLED (req))
1448 {
1449 errno = ECANCELED;
1450 break;
762 } 1451 }
763 } 1452 }
1453}
764 1454
765 if (errno) 1455#ifdef PAGESIZE
766 res = -1; 1456# define eio_pagesize() PAGESIZE
1457#else
1458static intptr_t
1459eio_pagesize (void)
1460{
1461 static intptr_t page;
1462
1463 if (!page)
1464 page = sysconf (_SC_PAGESIZE);
1465
1466 return page;
1467}
1468#endif
1469
1470static void
1471eio_page_align (void **addr, size_t *length)
1472{
1473 intptr_t mask = eio_pagesize () - 1;
1474
1475 /* round down addr */
1476 intptr_t adj = mask & (intptr_t)*addr;
1477
1478 *addr = (void *)((intptr_t)*addr - adj);
1479 *length += adj;
1480
1481 /* round up length */
1482 *length = (*length + mask) & ~mask;
1483}
1484
1485#if !_POSIX_MEMLOCK
1486# define eio__mlockall(a) ((errno = ENOSYS), -1)
1487#else
1488
1489static int
1490eio__mlockall (int flags)
1491{
1492 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1493 extern int mallopt (int, int);
1494 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1495 #endif
1496
1497 if (EIO_MCL_CURRENT != MCL_CURRENT
1498 || EIO_MCL_FUTURE != MCL_FUTURE)
1499 {
1500 flags = 0
1501 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1502 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1503 }
1504
1505 return mlockall (flags);
1506}
1507#endif
1508
1509#if !_POSIX_MEMLOCK_RANGE
1510# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1511#else
1512
1513static int
1514eio__mlock (void *addr, size_t length)
1515{
1516 eio_page_align (&addr, &length);
1517
1518 return mlock (addr, length);
1519}
1520
1521#endif
1522
1523#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1524# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1525#else
1526
1527int
1528eio__msync (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 if (EIO_MS_ASYNC != MS_SYNC
1533 || EIO_MS_INVALIDATE != MS_INVALIDATE
1534 || EIO_MS_SYNC != MS_SYNC)
1535 {
1536 flags = 0
1537 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1538 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1539 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1540 }
1541
1542 return msync (mem, len, flags);
1543}
1544
1545#endif
1546
1547int
1548eio__mtouch (eio_req *req)
1549{
1550 void *mem = req->ptr2;
1551 size_t len = req->size;
1552 int flags = req->int1;
1553
1554 eio_page_align (&mem, &len);
1555
767 1556 {
768 req->result = res; 1557 intptr_t addr = (intptr_t)mem;
1558 intptr_t end = addr + len;
1559 intptr_t page = eio_pagesize ();
1560
1561 if (addr < end)
1562 if (flags & EIO_MT_MODIFY) /* modify */
1563 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1564 else
1565 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1566 }
1567
1568 return 0;
769} 1569}
770 1570
771/*****************************************************************************/ 1571/*****************************************************************************/
772 1572
1573#define ALLOC(len) \
1574 if (!req->ptr2) \
1575 { \
1576 X_LOCK (wrklock); \
1577 req->flags |= EIO_FLAG_PTR2_FREE; \
1578 X_UNLOCK (wrklock); \
1579 req->ptr2 = malloc (len); \
1580 if (!req->ptr2) \
1581 { \
1582 errno = ENOMEM; \
1583 req->result = -1; \
1584 break; \
1585 } \
1586 }
1587
773X_THREAD_PROC (eio_proc) 1588X_THREAD_PROC (etp_proc)
774{ 1589{
775 eio_req *req; 1590 ETP_REQ *req;
776 struct timespec ts; 1591 struct timespec ts;
777 worker *self = (worker *)thr_arg; 1592 etp_worker *self = (etp_worker *)thr_arg;
778 1593
779 /* try to distribute timeouts somewhat randomly */ 1594 /* try to distribute timeouts somewhat randomly */
780 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1595 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
781 1596
782 for (;;) 1597 for (;;)
783 { 1598 {
784 ts.tv_sec = time (0) + IDLE_TIMEOUT;
785
786 X_LOCK (reqlock); 1599 X_LOCK (reqlock);
787 1600
788 for (;;) 1601 for (;;)
789 { 1602 {
790 self->req = req = reqq_shift (&req_queue); 1603 self->req = req = reqq_shift (&req_queue);
… …
792 if (req) 1605 if (req)
793 break; 1606 break;
794 1607
795 ++idle; 1608 ++idle;
796 1609
1610 ts.tv_sec = time (0) + idle_timeout;
797 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1611 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
798 { 1612 {
799 if (idle > max_idle) 1613 if (idle > max_idle)
800 { 1614 {
801 --idle; 1615 --idle;
… …
806 goto quit; 1620 goto quit;
807 } 1621 }
808 1622
809 /* we are allowed to idle, so do so without any timeout */ 1623 /* we are allowed to idle, so do so without any timeout */
810 X_COND_WAIT (reqwait, reqlock); 1624 X_COND_WAIT (reqwait, reqlock);
811 ts.tv_sec = time (0) + IDLE_TIMEOUT;
812 } 1625 }
813 1626
814 --idle; 1627 --idle;
815 } 1628 }
816 1629
817 --nready; 1630 --nready;
818 1631
819 X_UNLOCK (reqlock); 1632 X_UNLOCK (reqlock);
820 1633
821 errno = 0; /* strictly unnecessary */ 1634 if (req->type < 0)
1635 goto quit;
822 1636
823 if (!EIO_CANCELLED (req)) 1637 if (!EIO_CANCELLED (req))
824 switch (req->type) 1638 ETP_EXECUTE (self, req);
825 {
826 case EIO_READ: req->result = req->offs >= 0
827 ? pread (req->int1, req->ptr2, req->size, req->offs)
828 : read (req->int1, req->ptr2, req->size); break;
829 case EIO_WRITE: req->result = req->offs >= 0
830 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
831 : write (req->int1, req->ptr2, req->size); break;
832
833 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
834 case EIO_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break;
835
836 case EIO_STAT: req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
837 case EIO_LSTAT: req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
838 case EIO_FSTAT: req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
839
840 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
841 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
842 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
843 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
844 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
845 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
846
847 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
848 case EIO_CLOSE: req->result = close (req->int1); break;
849 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
850 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
851 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
852 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
853 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
854 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
855 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
856 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
857 case EIO_READLINK: req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
858
859 case EIO_SYNC: req->result = 0; sync (); break;
860 case EIO_FSYNC: req->result = fsync (req->int1); break;
861 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
862
863 case EIO_READDIR: scandir_ (req, self); break;
864
865 case EIO_BUSY:
866#ifdef _WIN32
867 Sleep (req->nv1 * 1000.);
868#else
869 {
870 struct timeval tv;
871
872 tv.tv_sec = req->nv1;
873 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
874
875 req->result = select (0, 0, 0, 0, &tv);
876 }
877#endif
878 break;
879
880 case EIO_UTIME:
881 case EIO_FUTIME:
882 {
883 struct timeval tv[2];
884 struct timeval *times;
885
886 if (req->nv1 != -1. || req->nv2 != -1.)
887 {
888 tv[0].tv_sec = req->nv1;
889 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
890 tv[1].tv_sec = req->nv2;
891 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
892
893 times = tv;
894 }
895 else
896 times = 0;
897
898
899 req->result = req->type == EIO_FUTIME
900 ? futimes (req->int1, times)
901 : utimes (req->ptr1, times);
902 }
903
904 case EIO_GROUP:
905 case EIO_NOP:
906 break;
907
908 case EIO_QUIT:
909 goto quit;
910
911 default:
912 req->result = -1;
913 break;
914 }
915
916 req->errorno = errno;
917 1639
918 X_LOCK (reslock); 1640 X_LOCK (reslock);
919 1641
920 ++npending; 1642 ++npending;
921 1643
922 if (!reqq_push (&res_queue, req) && want_poll_cb) 1644 if (!reqq_push (&res_queue, req) && want_poll_cb)
923 want_poll_cb (); 1645 want_poll_cb ();
924 1646
925 self->req = 0; 1647 self->req = 0;
926 worker_clear (self); 1648 etp_worker_clear (self);
927 1649
928 X_UNLOCK (reslock); 1650 X_UNLOCK (reslock);
929 } 1651 }
930 1652
931quit: 1653quit:
932 X_LOCK (wrklock); 1654 X_LOCK (wrklock);
933 worker_free (self); 1655 etp_worker_free (self);
934 X_UNLOCK (wrklock); 1656 X_UNLOCK (wrklock);
935 1657
936 return 0; 1658 return 0;
937} 1659}
938 1660
939/*****************************************************************************/ 1661/*****************************************************************************/
940 1662
941static void eio_atfork_prepare (void) 1663int eio_init (void (*want_poll)(void), void (*done_poll)(void))
942{ 1664{
943 X_LOCK (wrklock); 1665 return etp_init (want_poll, done_poll);
944 X_LOCK (reqlock);
945 X_LOCK (reslock);
946#if !HAVE_PREADWRITE
947 X_LOCK (preadwritelock);
948#endif
949#if !HAVE_READDIR_R
950 X_LOCK (readdirlock);
951#endif
952} 1666}
953 1667
954static void eio_atfork_parent (void) 1668static void eio_api_destroy (eio_req *req)
955{ 1669{
956#if !HAVE_READDIR_R 1670 free (req);
957 X_UNLOCK (readdirlock);
958#endif
959#if !HAVE_PREADWRITE
960 X_UNLOCK (preadwritelock);
961#endif
962 X_UNLOCK (reslock);
963 X_UNLOCK (reqlock);
964 X_UNLOCK (wrklock);
965} 1671}
966 1672
967static void eio_atfork_child (void) 1673#define REQ(rtype) \
968{ 1674 eio_req *req; \
969 eio_req *prv; 1675 \
1676 req = (eio_req *)calloc (1, sizeof *req); \
1677 if (!req) \
1678 return 0; \
1679 \
1680 req->type = rtype; \
1681 req->pri = pri; \
1682 req->finish = cb; \
1683 req->data = data; \
1684 req->destroy = eio_api_destroy;
970 1685
971 while (prv = reqq_shift (&req_queue)) 1686#define SEND eio_submit (req); return req
972 eio_destroy (prv);
973 1687
974 while (prv = reqq_shift (&res_queue)) 1688#define PATH \
975 eio_destroy (prv); 1689 req->flags |= EIO_FLAG_PTR1_FREE; \
1690 req->ptr1 = strdup (path); \
1691 if (!req->ptr1) \
1692 { \
1693 eio_api_destroy (req); \
1694 return 0; \
1695 }
976 1696
977 while (wrk_first.next != &wrk_first) 1697static void eio_execute (etp_worker *self, eio_req *req)
1698{
1699 switch (req->type)
978 { 1700 {
979 worker *wrk = wrk_first.next; 1701 case EIO_READ: ALLOC (req->size);
1702 req->result = req->offs >= 0
1703 ? pread (req->int1, req->ptr2, req->size, req->offs)
1704 : read (req->int1, req->ptr2, req->size); break;
1705 case EIO_WRITE: req->result = req->offs >= 0
1706 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1707 : write (req->int1, req->ptr2, req->size); break;
980 1708
981 if (wrk->req) 1709 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
982 eio_destroy (wrk->req); 1710 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
983 1711
984 worker_clear (wrk); 1712 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
985 worker_free (wrk); 1713 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
986 } 1714 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1715 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1716 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1717 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
987 1718
988 started = 0; 1719 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
989 idle = 0; 1720 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
990 nreqs = 0; 1721 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
991 nready = 0; 1722 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
992 npending = 0;
993 1723
994 eio_atfork_parent (); 1724 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
995} 1725 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1726 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1727 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1728 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1729 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
996 1730
997int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1731 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
998{ 1732 case EIO_CLOSE: req->result = close (req->int1); break;
999 want_poll_cb = want_poll; 1733 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1000 done_poll_cb = done_poll; 1734 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1735 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1736 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1737 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1738 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1739 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1740 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1001 1741
1742 case EIO_READLINK: ALLOC (PATH_MAX);
1743 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1744
1745 case EIO_SYNC: req->result = 0; sync (); break;
1746 case EIO_FSYNC: req->result = fsync (req->int1); break;
1747 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1748 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1749 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1750 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1751 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1752 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1753
1754 case EIO_READDIR: eio__scandir (req, self); break;
1755
1756 case EIO_BUSY:
1002#ifdef _WIN32 1757#ifdef _WIN32
1003 X_MUTEX_CHECK (wrklock); 1758 Sleep (req->nv1 * 1e3);
1004 X_MUTEX_CHECK (reslock); 1759#else
1005 X_MUTEX_CHECK (reqlock);
1006 X_MUTEX_CHECK (reqwait);
1007 X_MUTEX_CHECK (preadwritelock);
1008 X_MUTEX_CHECK (readdirlock);
1009
1010 X_COND_CHECK (reqwait);
1011#endif
1012
1013 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1014}
1015
1016#if 0
1017
1018eio_req *eio_fsync (int fd, eio_cb cb);
1019eio_req *eio_fdatasync (int fd, eio_cb cb);
1020eio_req *eio_dupclose (int fd, eio_cb cb);
1021eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb);
1022eio_req *eio_read (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1023eio_req *eio_write (int fd, off_t offs, size_t length, char *data, eio_cb cb);
1024eio_req *eio_fstat (int fd, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1025eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb);
1026eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb);
1027eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb);
1028eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb);
1029eio_req *eio_dup2 (int fd, int fd2, eio_cb cb);
1030eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb);
1031eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb);
1032eio_req *eio_readlink (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1033eio_req *eio_stat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1034eio_req *eio_lstat (const char *path, eio_cb cb); /* stat buffer=ptr2 allocates dynamically */
1035eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb);
1036eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb);
1037eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb);
1038eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb);
1039eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb);
1040eio_req *eio_unlink (const char *path, eio_cb cb);
1041eio_req *eio_rmdir (const char *path, eio_cb cb);
1042eio_req *eio_readdir (const char *path, eio_cb cb); /* result=ptr2 allocated dynamically */
1043eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb);
1044eio_req *eio_busy (double delay, eio_cb cb); /* ties a thread for this long, simulating busyness */
1045eio_req *eio_nop (eio_cb cb); /* does nothing except go through the whole process */
1046void
1047aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef)
1048 PROTOTYPE: $$$;$
1049 PPCODE:
1050{
1051 dREQ;
1052
1053 req->type = EIO_OPEN;
1054 req->sv1 = newSVsv (pathname);
1055 req->ptr1 = SvPVbyte_nolen (req->sv1);
1056 req->int1 = flags;
1057 req->int2 = mode;
1058
1059 EIO_SEND;
1060}
1061
1062void
1063aio_fsync (SV *fh, SV *callback=&PL_sv_undef)
1064 PROTOTYPE: $;$
1065 ALIAS:
1066 aio_fsync = EIO_FSYNC
1067 aio_fdatasync = EIO_FDATASYNC
1068 PPCODE:
1069{
1070 dREQ;
1071
1072 req->type = ix;
1073 req->sv1 = newSVsv (fh);
1074 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1075
1076 EIO_SEND (req);
1077}
1078
1079void
1080aio_close (SV *fh, SV *callback=&PL_sv_undef)
1081 PROTOTYPE: $;$
1082 PPCODE:
1083{
1084 dREQ;
1085
1086 req->type = EIO_CLOSE;
1087 req->sv1 = newSVsv (fh);
1088 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1089
1090 EIO_SEND (req);
1091}
1092
1093void
1094aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef)
1095 ALIAS:
1096 aio_read = EIO_READ
1097 aio_write = EIO_WRITE
1098 PROTOTYPE: $$$$$;$
1099 PPCODE:
1100{
1101 STRLEN svlen;
1102 char *svptr = SvPVbyte (data, svlen);
1103 UV len = SvUV (length);
1104
1105 SvUPGRADE (data, SVt_PV);
1106 SvPOK_on (data);
1107
1108 if (dataoffset < 0)
1109 dataoffset += svlen;
1110
1111 if (dataoffset < 0 || dataoffset > svlen)
1112 croak ("dataoffset outside of data scalar");
1113
1114 if (ix == EIO_WRITE)
1115 {
1116 /* write: check length and adjust. */
1117 if (!SvOK (length) || len + dataoffset > svlen)
1118 len = svlen - dataoffset;
1119 }
1120 else
1121 {
1122 /* read: grow scalar as necessary */
1123 svptr = SvGROW (data, len + dataoffset + 1);
1124 }
1125
1126 if (len < 0)
1127 croak ("length must not be negative");
1128
1129 { 1760 {
1130 dREQ; 1761 struct timeval tv;
1131 1762
1132 req->type = ix; 1763 tv.tv_sec = req->nv1;
1133 req->sv1 = newSVsv (fh); 1764 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1134 req->int1 = PerlIO_fileno (ix == EIO_READ ? IoIFP (sv_2io (fh))
1135 : IoOFP (sv_2io (fh)));
1136 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1137 req->size = len;
1138 req->sv2 = SvREFCNT_inc (data);
1139 req->ptr2 = (char *)svptr + dataoffset;
1140 req->stroffset = dataoffset;
1141 1765
1142 if (!SvREADONLY (data)) 1766 req->result = select (0, 0, 0, 0, &tv);
1767 }
1768#endif
1769 break;
1770
1771 case EIO_UTIME:
1772 case EIO_FUTIME:
1773 {
1774 struct timeval tv[2];
1775 struct timeval *times;
1776
1777 if (req->nv1 != -1. || req->nv2 != -1.)
1143 { 1778 {
1144 SvREADONLY_on (data); 1779 tv[0].tv_sec = req->nv1;
1145 req->flags |= FLAG_SV2_RO_OFF; 1780 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1781 tv[1].tv_sec = req->nv2;
1782 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1783
1784 times = tv;
1146 } 1785 }
1786 else
1787 times = 0;
1147 1788
1148 EIO_SEND; 1789 req->result = req->type == EIO_FUTIME
1790 ? futimes (req->int1, times)
1791 : utimes (req->ptr1, times);
1149 } 1792 }
1150} 1793 break;
1151 1794
1152void 1795 case EIO_GROUP:
1153aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) 1796 abort (); /* handled in eio_request */
1154 PROTOTYPE: $$;$
1155 PPCODE:
1156{
1157 SV *data;
1158 dREQ;
1159 1797
1160 data = newSV (NAME_MAX); 1798 case EIO_NOP:
1161 SvPOK_on (data); 1799 req->result = 0;
1800 break;
1162 1801
1163 req->type = EIO_READLINK; 1802 case EIO_CUSTOM:
1164 req->sv1 = newSVsv (path); 1803 req->feed (req);
1165 req->ptr1 = SvPVbyte_nolen (req->sv1); 1804 break;
1166 req->sv2 = data;
1167 req->ptr2 = SvPVbyte_nolen (data);
1168 1805
1169 EIO_SEND; 1806 default:
1170} 1807 errno = ENOSYS;
1808 req->result = -1;
1809 break;
1810 }
1171 1811
1172void 1812 req->errorno = errno;
1173aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef)
1174 PROTOTYPE: $$$$;$
1175 PPCODE:
1176{
1177 dREQ;
1178
1179 req->type = EIO_SENDFILE;
1180 req->sv1 = newSVsv (out_fh);
1181 req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1182 req->sv2 = newSVsv (in_fh);
1183 req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1184 req->offs = SvVAL64 (in_offset);
1185 req->size = length;
1186
1187 EIO_SEND;
1188} 1813}
1189 1814
1190void 1815#ifndef EIO_NO_WRAPPERS
1191aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef)
1192 PROTOTYPE: $$$;$
1193 PPCODE:
1194{
1195 dREQ;
1196 1816
1197 req->type = EIO_READAHEAD; 1817eio_req *eio_nop (int pri, eio_cb cb, void *data)
1198 req->sv1 = newSVsv (fh);
1199 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh)));
1200 req->offs = SvVAL64 (offset);
1201 req->size = length;
1202
1203 EIO_SEND;
1204}
1205
1206void
1207aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef)
1208 ALIAS:
1209 aio_stat = EIO_STAT
1210 aio_lstat = EIO_LSTAT
1211 PPCODE:
1212{ 1818{
1213 dREQ; 1819 REQ (EIO_NOP); SEND;
1820}
1214 1821
1215 req->ptr2 = malloc (sizeof (EIO_STRUCT_STAT)); 1822eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1216 if (!req->ptr2) 1823{
1217 { 1824 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1218 req_destroy (req); 1825}
1219 croak ("out of memory during aio_stat statdata allocation");
1220 }
1221 1826
1827eio_req *eio_sync (int pri, eio_cb cb, void *data)
1828{
1829 REQ (EIO_SYNC); SEND;
1830}
1831
1832eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1833{
1834 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1835}
1836
1837eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1838{
1839 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1840}
1841
1842eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1843{
1844 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1845}
1846
1847eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
1848{
1849 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
1850}
1851
1852eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1853{
1854 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1855}
1856
1857eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1858{
1859 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1860}
1861
1862eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1863{
1864 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1865}
1866
1867eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1868{
1869 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1870}
1871
1872eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1873{
1874 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1875}
1876
1877eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1878{
1879 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1880}
1881
1882eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1883{
1884 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1885}
1886
1887eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1888{
1889 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1890}
1891
1892eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
1893{
1894 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1895}
1896
1897eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1898{
1899 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1900}
1901
1902eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1903{
1904 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1905}
1906
1907eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1908{
1909 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1910}
1911
1912eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1913{
1914 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1915}
1916
1917eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1918{
1919 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1920}
1921
1922eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1923{
1924 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1925}
1926
1927eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1928{
1929 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1930}
1931
1932eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1933{
1934 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1935}
1936
1937eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1938{
1939 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1940}
1941
1942eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data)
1943{
1944 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1945}
1946
1947eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1948{
1949 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1950}
1951
1952eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1953{
1954 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1955}
1956
1957static eio_req *
1958eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1959{
1960 REQ (type); PATH; SEND;
1961}
1962
1963eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1964{
1965 return eio__1path (EIO_READLINK, path, pri, cb, data);
1966}
1967
1968eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1969{
1970 return eio__1path (EIO_STAT, path, pri, cb, data);
1971}
1972
1973eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1974{
1975 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1976}
1977
1978eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
1979{
1980 return eio__1path (EIO_STATVFS, path, pri, cb, data);
1981}
1982
1983eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1984{
1985 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1986}
1987
1988eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1989{
1990 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1991}
1992
1993eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1994{
1995 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1996}
1997
1998eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1999{
2000 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
2001}
2002
2003static eio_req *
2004eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2005{
2006 REQ (type); PATH;
2007
1222 req->flags |= FLAG_PTR2_FREE; 2008 req->flags |= EIO_FLAG_PTR2_FREE;
1223 req->sv1 = newSVsv (fh_or_path); 2009 req->ptr2 = strdup (new_path);
2010 if (!req->ptr2)
2011 {
2012 eio_api_destroy (req);
2013 return 0;
2014 }
1224 2015
1225 if (SvPOK (fh_or_path)) 2016 SEND;
1226 {
1227 req->type = ix;
1228 req->ptr1 = SvPVbyte_nolen (req->sv1);
1229 }
1230 else
1231 {
1232 req->type = EIO_FSTAT;
1233 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1234 }
1235
1236 EIO_SEND;
1237} 2017}
1238 2018
1239void 2019eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1240aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef)
1241 PPCODE:
1242{ 2020{
1243 dREQ; 2021 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1244
1245 req->nv1 = SvOK (atime) ? SvNV (atime) : -1.;
1246 req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.;
1247 req->sv1 = newSVsv (fh_or_path);
1248
1249 if (SvPOK (fh_or_path))
1250 {
1251 req->type = EIO_UTIME;
1252 req->ptr1 = SvPVbyte_nolen (req->sv1);
1253 }
1254 else
1255 {
1256 req->type = EIO_FUTIME;
1257 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1258 }
1259
1260 EIO_SEND;
1261} 2022}
1262 2023
1263void 2024eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1264aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef)
1265 PPCODE:
1266{ 2025{
1267 dREQ; 2026 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1268
1269 req->sv1 = newSVsv (fh_or_path);
1270 req->offs = SvOK (offset) ? SvVAL64 (offset) : -1;
1271
1272 if (SvPOK (fh_or_path))
1273 {
1274 req->type = EIO_TRUNCATE;
1275 req->ptr1 = SvPVbyte_nolen (req->sv1);
1276 }
1277 else
1278 {
1279 req->type = EIO_FTRUNCATE;
1280 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1281 }
1282
1283 EIO_SEND;
1284} 2027}
1285 2028
1286void 2029eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1287aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef)
1288 ALIAS:
1289 aio_chmod = EIO_CHMOD
1290 aio_fchmod = EIO_FCHMOD
1291 aio_mkdir = EIO_MKDIR
1292 PPCODE:
1293{ 2030{
1294 dREQ; 2031 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1295
1296 req->type = type;
1297 req->int2 = mode;
1298 req->sv1 = newSVsv (fh_or_path);
1299
1300 if (ix == EIO_FCHMOD)
1301 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1302 else
1303 req->ptr1 = SvPVbyte_nolen (req->sv1);
1304
1305 EIO_SEND;
1306} 2032}
1307 2033
1308void 2034eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1309aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef)
1310 PPCODE:
1311{ 2035{
1312 dREQ; 2036 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1313
1314 req->int2 = SvOK (uid) ? SvIV (uid) : -1;
1315 req->int3 = SvOK (gid) ? SvIV (gid) : -1;
1316 req->sv1 = newSVsv (fh_or_path);
1317
1318 if (SvPOK (fh_or_path))
1319 {
1320 req->type = EIO_CHOWN;
1321 req->ptr1 = SvPVbyte_nolen (req->sv1);
1322 }
1323 else
1324 {
1325 req->type = EIO_FCHOWN;
1326 req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1327 }
1328
1329 EIO_SEND;
1330} 2037}
1331 2038
1332void
1333aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef)
1334 ALIAS:
1335 aio_unlink = EIO_UNLINK
1336 aio_rmdir = EIO_RMDIR
1337 aio_readdir = EIO_READDIR
1338 PPCODE:
1339{
1340 dREQ;
1341
1342 req->type = ix;
1343 req->sv1 = newSVsv (pathname);
1344 req->ptr1 = SvPVbyte_nolen (req->sv1);
1345
1346 EIO_SEND;
1347}
1348
1349void
1350aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef)
1351 ALIAS:
1352 aio_link = EIO_LINK
1353 aio_symlink = EIO_SYMLINK
1354 aio_rename = EIO_RENAME
1355 PPCODE:
1356{
1357 dREQ;
1358
1359 req->type = ix;
1360 req->sv1 = newSVsv (oldpath);
1361 req->ptr1 = SvPVbyte_nolen (req->sv1);
1362 req->sv2 = newSVsv (newpath);
1363 req->ptr2 = SvPVbyte_nolen (req->sv2);
1364
1365 EIO_SEND;
1366}
1367
1368void
1369aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef)
1370 PPCODE:
1371{
1372 dREQ;
1373
1374 req->type = EIO_MKNOD;
1375 req->sv1 = newSVsv (pathname);
1376 req->ptr1 = SvPVbyte_nolen (req->sv1);
1377 req->int2 = (mode_t)mode;
1378 req->offs = dev;
1379
1380 EIO_SEND;
1381}
1382
1383void
1384aio_busy (double delay, SV *callback=&PL_sv_undef)
1385 PPCODE:
1386{
1387 dREQ;
1388
1389 req->type = EIO_BUSY;
1390 req->nv1 = delay < 0. ? 0. : delay;
1391
1392 EIO_SEND;
1393}
1394
1395void
1396aio_group (SV *callback=&PL_sv_undef)
1397 PROTOTYPE: ;$
1398 PPCODE:
1399{
1400 dREQ;
1401
1402 req->type = EIO_GROUP;
1403
1404 req_send (req);
1405 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1406}
1407
1408void
1409aio_nop (SV *callback=&PL_sv_undef)
1410 ALIAS:
1411 aio_nop = EIO_NOP
1412 aio_sync = EIO_SYNC
1413 PPCODE:
1414{
1415 dREQ;
1416
1417#endif 2039#endif
2040
2041eio_req *eio_grp (eio_cb cb, void *data)
2042{
2043 const int pri = EIO_PRI_MAX;
2044
2045 REQ (EIO_GROUP); SEND;
2046}
2047
2048#undef REQ
2049#undef PATH
2050#undef SEND
2051
2052/*****************************************************************************/
2053/* grp functions */
1418 2054
1419void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2055void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1420{ 2056{
1421 grp->int2 = limit; 2057 grp->int2 = limit;
1422 grp->feed = feed; 2058 grp->feed = feed;
… …
1433 2069
1434void eio_grp_add (eio_req *grp, eio_req *req) 2070void eio_grp_add (eio_req *grp, eio_req *req)
1435{ 2071{
1436 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2072 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1437 2073
2074 grp->flags |= EIO_FLAG_GROUPADD;
2075
1438 ++grp->size; 2076 ++grp->size;
1439 req->grp = grp; 2077 req->grp = grp;
1440 2078
1441 req->grp_prev = 0; 2079 req->grp_prev = 0;
1442 req->grp_next = grp->grp_first; 2080 req->grp_next = grp->grp_first;
… …
1445 grp->grp_first->grp_prev = req; 2083 grp->grp_first->grp_prev = req;
1446 2084
1447 grp->grp_first = req; 2085 grp->grp_first = req;
1448} 2086}
1449 2087
2088/*****************************************************************************/
2089/* misc garbage */
1450 2090
2091ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2092{
2093 etp_worker wrk;
2094 ssize_t ret;
2095
2096 wrk.dbuf = 0;
2097
2098 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2099
2100 if (wrk.dbuf)
2101 free (wrk.dbuf);
2102
2103 return ret;
2104}
2105

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines